Leveraging co-registration in T1/T2/T2*/fat-fraction (FF) MR Multitasking to predict spectroscopic FF in pre-clinical HFpEF (RF_SA_460)

Adipose tissue plays a key role in cardiovascular disease progression, so monitoring its changes may help detect pathological processes1. MR spectroscopy (MRS) is the gold standard for fat fraction (FF) quantification2 but is burdensome and limited to a single voxel. Dixon-based chemical shift MRI is a fast, spatially-resolved alternative3, but its FF precision suffers from system imperfections and difficulty pinpointing exact lipid resonance frequencies. Studies show limits of agreement (LOA) from –4.44 to 6.75% between Dixon-based techniques and MRS3, hindering Dixon-based detection of small changes in adipose content. For instance, a recent study of heart failure with preserved ejection fraction (HFpEF) in female patients identified a need to detect a 0.4% difference in FF4.

This study evaluates whether T1-T2-T2*-FF mapping with CMR Multitasking5 can bridge this gap by using the co-registered T1, T2, and/or T2* maps to account for the discrepancy in FF values in sub-clinical HFpEF patients.

Methods:

In-Vivo Study

N = 10 sub-clinical HFpEF patients were scanned on a 3T Siemens Vida scanner. A 2.5-min CMR Multitasking mid-ventricular short-axis scan measured T1, T2, T2*, and FF. A free-breathing single-voxel MRS protocol was performed in the mid-septum during systole. Fat-fraction values FFMRS and FFMT were extracted from the MRS and Multitasking acquisitions, respectively, using the systolic FFMT phase to match MRS. The root-mean-square error (RMSE) between FFMRS and bias-corrected FFMT was calculated as a baseline for comparison. Potential limitations include differences in spatial resolutions for FF_MT and FF_MRS, as well as differences in respiratory motion states.

Initial Biomarker Screening

To determine which biomarkers—of T1, T2, and T2*—best explain the discrepancy between imaging and MRS, we performed linear regression between each biomarker and the FF estimation residual (FF_MRS–FF_MT), across all subjects. Biomarkers with statistically significant (p< 0.05) Pearson correlation were selected for follow-up study.

Leave-one-out (LOO) validation
The only biomarker selected for follow-up validation was T2, as described in results. The linear regression model using T2 as input was retrained using LOO validation. Agreement between FF_MRS and FF_MT with and without LOO T2 correction was analyzed using Bland–Altman plots.

Results: Example images are shown in Fig. 1. One subject showed large streaking artifacts in the FF map and was excluded from analysis. The baseline RMSE between FFMRS and FFMT was 2.42%. In the screening regressions (Fig. 2), only T2 significantly correlated with the FF residual (p =0.008). LOO validation of T2 as an additional fat predictor halved RMSE, to 1.12%. Fig. 3 compares Bland-Altman plots. T2 improved LOA from –2.03% ± 2.74% to –0.13% ± 2.06%.

Conclusion: Complementary, co-registered parameter maps from MR Multitasking can improve FF agreement versus MRS. In this cohort, T2 was the only relevant biomarker; further study is needed to see if this correction is accounting for T2-weighting of spin-echo FFMRS estimates or reducing T2-confounding of FFMT. Larger studies would allow for multivariate analysis beyond linear regression.